The agency of one, on a single screen.

An agency is usually a pile of people. I'm testing whether it can be a pile of services instead — each one owning a single outcome for a single client, run by AI workers, with me as the only human in the loop. The diagram below is that system's operating picture. It plays on its own, and you can also walk it one step at a time.

It isn't a mockup. It's the actual model the work runs on, drawn honestly — including the part where a machine has to stop and ask me before it spends your money or touches the outside world.

How to read it

1. The Service sits at the center. It's the durable unit of the firm — one outcome, one owner. Every tool, model, and vendor around it is replaceable. The service, and what it has learned, is the asset.
2. The Client World Model wraps it. One protected, current understanding of each client — goals, permissions, hard-won facts. It informs every decision. It never grants one. Context, not authority.
3. A spine of five questions runs every task: what is true (KNOW), what is possible (CAN), what is permitted (MAY), what happens now (DO), and what should change (LEARN).
4. Watch the gate at MAY. Every signal that walks the loop stops there. Routine, proven, in-bounds moves pass on their own. Anything that spends real money or reaches a client halts and waits for a yes. Being capable never counts as being allowed.

Pick a scenario — a paid-search money leak, an SEO clue, a quiet customer about to churn — and watch the same skeleton handle all three. It auto-plays; click the dots to move at your own pace.

The tools are named by role on purpose — the senses, the specialists, the long memory — never by brand. Swap the vendor, keep the system. That's the whole point: the moat isn't any one model, it's the service and the rule it operates under.